Introduction
Cardiac magnetic resonance imaging (MRI) provides incremental impact on clinical decision-making by delivering data about myocardial anatomy, function, tissue characterization, perfusion, diffusion and also metabolism when using MR spectroscopy [
1]. MRI is the clinical standard of reference for the volumetric and functional assessment of the heart [
2‐
5]. Cardiac positron emission tomography (PET) can provide precise information on cardiac perfusion, viability, myocardial metabolism and other molecular processes [
6]. [
18F] FDG cardiac PET visualizes the glucose metabolism in the viable myocyte and is the gold standard for the direct visualization of viability [
7]. The unique combination of PET and MRI offers a comprehensive approach for cardiac diseases [
6,
8], i.e. in inflammatory diseases like myocarditis, sarcoidosis or rejection after heart transplantation [
9,
10]. Both modalities can assess cardiac function.
Recently, hybrid PET/MR systems have been introduced [
11], and allow the evaluation of simultaneously acquired volumetric cardiac data of both modalities. Thus, the question arises whether volumetric parameters could be used interchangeably. Validation studies on co-registering of sequentially acquired [
18F] FDG-PET and MRI data sets showed a good agreement of LV parameters with a systematic bias [
12‐
14]. However, inter- and/or intraday variability of LV parameters can occur, if not acquired simultaneously [
14]. Even the inherent differences in acquisition times between PET and MRI can cause misalignment, mostly due to patient motion or respiration [
8].
LV function and especially the normalized LVESV (LVESVI) is a relevant prognostic factor in patients with coronary artery disease (CAD) and a LVESVI > 100 ml/m
2 predicts worse outcome [
15]. In comparative studies, [
18F] FDG-PET showed a non-significant tendency to overestimate LVESV and to underestimate LVEDV, resulting in an underestimation of LVEF values [
2]. The lower temporal and spatial resolution of cardiac [
18F] FDG-PET [
2] as compared to MRI [
8] might be a reason. But, the temporal resolution of gated [
18F] FDG-PET has been improved lately and spatial resolution of MRI is lower in the through-plane direction. Also, cardiac function by [
18F] FDG-PET has been mainly evaluated in patients with CAD, whom often show a heterogeneous myocardial uptake, while patients who undergo [
18F] FDG-PET for non-cardiac reasons sometimes show a homogenous uptake [
16].
The aims of this study were:
1.
To test whether an observed homogeneous uptake is sufficient to perform a simultaneous cardiac PET/MR with the residual activity after a PET/CT for non-cardiac indication.
2.
To analyse whether the acquired data is sufficient for a LV volumetric and functional analysis.
3.
To assess whether bias and limits of agreement are small enough in a way that cardiac PET or MRI volumetric and functional data could be used interchangeably.
Discussion
Assessment of left ventricular volumes, function and mass data is feasible with a homogenous residual activity of a previous PET/CT examination. If acquired with simultaneous cardiac PET/MR, parameters correlated well between PET and MRI readouts with only small bias, (Table
3). However, the limits of agreement (LOAs) are rather wide, such that PET and MRI functional and volumetric data could be used interchangeably.
Intra- and interobserver reliability was slightly lower in PET as compared to MRI, although MRI is a frequently used standard of reference. These results are not surprising, because the cardiac PET algorithm for the volumetric and functional evaluation does not allow much user-software interaction. Observers can only change the LV axis, the position of the apex and the mitral valve. Epicardial and endocardial contours are detected automatically and cannot be changed manually. Papillary muscles are merged into the contours or neglected due to low uptake. With more possible user interaction, the inter- and intraobserver variability increases [
25]. The contours, defined in the MRI datasets, can be modified substantially and partial volume effects of the papillary muscles and the manual correction can lead to discrepancies in the different measurements. Therefore, we conclude that the higher intra- and interobserver variability deduces itself from higher user-software interaction and can be decreased by reader consensus training [
26].
We acquired the SSFP sequences in expiration, because our own unpublished experience in consensus to other studies has revealed [
20,
21] that this method leads to the most accurate thoracic image fusion, due to free breathing being mainly an end-expiration phase [
27]. Additional respiratory gating of the PET acquisition, usually triggered by MR navigator pulses, may improve image fusion, however, at the cost of higher acquisition time [
21,
27].
The correlation between MRI and PET with corresponding coefficients for LVEDV, LVESV and LVEF is comparable to other studies [
2,
12,
13], except LVMM, which was lower in one study [
13]. All three studies quoted in Table
3 examined patients exclusively with coronary CAD and reduced LVEF. In our study, only 31% of patients had CAD.
Different temporal resolutions can significantly alter the correct assessment of LVEF, if the end-diastolic and end-systolic phases are not detected properly [
28,
29]. In patients with severely impaired LVEF, the differences between LV volumes in each phase are smaller. Therefore, a lower temporal resolution has less influence on LVEF compared to patients with normal or slightly decreased LVEF [
4].
The TR of our used SSFP sequence was automatically adjusted to HR and ranged between 46–52 ms. The mean TR of our examined cohort was 50 ms (SD: ±1.3; range: 46–52), the mean HR was 64 bpm (SD: ±13; range: 41–94 bpm), the resulting mean RR interval was 970 ms (SD: ±196; range: 640–1475 ms). Therefore, in our cohort, the mean HR was rather low, and the corresponding RR interval was high, which are rather optimal conditions to achieve a high temporal resolution. Taking this information into account, we had a mean maximum phase duration of 970/20 = 48.5 ms to catch the systolic phase appropriately within 5% of the RR interval. However, a “temporal partial volume effect” may occur with a loss of precision in the determination of systolic volumes in some patients, and thus a potential underestimation of the ejection fraction.
Due to the fact that a fixed number of 16 gates were reconstructed with the PET acquisition, the mean gate length of the PET reconstruction was 60 ms (SD: ±11; range: 40–77 ms) in our cohort. Therefore, the mean temporal resolution of the PET acquisition was obviously lower than the temporal resolution of the MR acquisition with a mean percentage of the PET gate length with the MR TR of 118% (SD: ±23; range: 79–159%).
However, the reconstructed temporal resolution of MRI in our study was nearly 40% higher with a fixed number of 25 reconstructed phases per cardiac cycle in MRI as compared to a fixed number of 16 gates in cardiac PET, compared to all other cited studies (Table
2): Khorsand et al. (MRI 12–16 phases, PET: 8 gates), Schaefer et al. (MRI: 12–16 phases, PET: 8 gates) and Slart et al. (MRI: approximately 20 phases, PET: 16 gates). This difference in temporal resolution may not have significantly contributed to the differences of LV parameters between the methods, but may have contributed to the differences compared to other studies. Slart et al. had the best correlation for LVEF with an equal temporal resolution.
While the spatial “in-plane” resolution—the voxel size was 1.4 × 1.4× 8.0 mm3 for MRI and 1.4 × 1.4 × 2.03 mm3 for PET—was the same for MRI and PET, the spatial “through-plane” resolution was higher in PET. However, since the orientation in cardiac MRI is along the cardiac axis, while in PET it is along the z-axis of the patient, the spatial resolution is not directly comparable, especially if the variable position of the heart is considered. However, the higher spatial “through-plane” resolution of PET is, in our opinion, not the main factor for the difference between the results.
Comparable to other trials, the volumetric assessment of simultaneously acquired PET and MRI data showed significantly lower mean left ventricular volumes with PET as compared to MRI, also in the subgroups with and without CAD (Table
3). Furthermore, there were tendencies towards higher calculated LVEF and LVMM values with PET.
Non-CAD patients demonstrated similar results to Schaefer et al. [
2], but narrower LOA and a smaller bias for LVESV as well as LVEDV compared to other studies [
12,
13] and comparable results to another. LVEF showed similar bias and wider LOA [
12,
13]. However, the interobserver reliability in PET is lower than in the study of Khorsand et al. (values in brackets): −0.3 ±9 mL (5 ±16 mL) for LVEDV, 2% ±5% (1 ±5%) for LVEF and −3.3 ±12 g (24 ±17 g) for LVMM, although all values did not significantly differ from 0 [
12,
13].
In contrast to Khorsand et al., in our study, gated PET overestimated the mean LVMM by 18 g as compared to MRI. Also, different definitions of the LV cavity may contribute to this: In MRI, segmentation is driven by the visible anatomical structures, while in PET, algorithms of the maximal [
18F] FDG uptake in the myocardial wall are used to define an intramyocardial centreline from which the endo- and epicardial contours are estimated [
12]. Unfortunately, the algorithms behind these volumetric calculations are not fully public yet. Morphological data from simultaneously acquired MRI could be used to further improve segmentation algorithms for gated PET.
There is a lack of consensus concerning preparation of patients undergoing cardiac [
18F] FDG-PET [
16]. The nine patients for viability assessment received a glucose load [
11]. In contrast, the 20 patients that underwent PET/MR as a subsequent examination after PET/CT followed a fasting protocol that suppresses body muscle [
18F] FDG uptake but does not prevent an incidental homogenous cardiac uptake [
30]. In our case, only patients with this incidental homogenous myocardial uptake after PET/CT and who agreed to undergo another PET/MRI examination and had no contraindications to MRI were included in the study. This incidental homogenous uptake occurred in approx. 20% of the patients examined, which is in line with a prior study [
31]. Better image quality of the cardiac PET in patients that underwent PET/CT for non-cardiac reasons compared to that of patients with ischaemic scar tissue is probably due to the fact that the first group had presumably no history of cardiac pathologies (Table
3, Fig.
2).
However, differences in PET IQ influenced negatively the evaluation of LV parameters for LVMM calculations only (Table
2). The question of how the PET/MRI outcome parameter agreement as obtained in this feasibility study employing [18F] FDG as a viable PET tracer would translate into simultaneous PET/MRI studies utilizing blood flow PET tracers like [13N] ammonia or [83Rb] rubidium needs to be answered by subsequent studies. This especially refers to the LV function assessment in low-EF ranges in which the limited temporal resolution by the 8-bin PET gating techniques can lead to a bias.
Limitations
If our mean reported TR of 50 ms is considered as the “real” temporal resolution of our CINE-SSFP sequence, as some authors recommend [
32], rather than the reconstructed frames/phases per heartbeat, which were fixed at 25 phases, then the underestimation of PET functional parameters is not only caused by the use of residual activity, but also intrinsically due to the MR technology used itself. Furthermore, the interpretation of our results is limited due to the small number of patients, though comparable to other studies in terms of patient number and results. We examined patients with and without CAD, different to Slart, Schaefer, and Khorsand et al., who evaluated patients with known CAD and reduced LVEF. Due to the sequential study design, the average time between administration of [
18F] FDG and PET/MR was 2.3 h ±1.2 h, resulting in a lower count rate as compared to a standard [
18F] FDG-PET, which is performed 40–60 min after [
18F] FDG administration [
12]. However, it has been demonstrated, that the use of half of the usual activity in PET/MR seems prudent [
33] and high-definition hybrid cardiac FDG PET/MR has been shown to be diagnostic using a mean activity of only 150 ±70 MBq [
34], which is approximately half of the mean dose used in our study. It is comparable to the mean activity still available after 2.8 h on average in our non-CAD patients, which corresponds to 1.5 half-lifes of [
18F] FDG or 165 min.